Optimal policy for attention-modulated decisions explains human fixation behavior

Abstract

Traditional accumulation-to-bound decision-making models assume that all choice options are processed simultaneously with equal attention. In real life decisions, however, humans tend to alternate their visual fixation between individual items in order to efficiently gather relevant information [46, 23, 21, 12, 15]. These fixations also causally affect one’s choices, biasing them toward the longer-fixated item [38, 2, 25]. We derive a normative decision-making model in which fixating a choice item boosts information about that item. In contrast to previous models [25, 39], we assume that attention enhances the reliability of information rather than its magnitude, consistent with neurophysiological findings [3, 13, 29, 45]. Furthermore, our model actively controls fixation changes to optimize information gathering. We show that the optimal model reproduces fixation patterns and fixation-related choice biases seen in human decision-makers, and provides a Bayesian computational rationale for the fixation bias. This insight led to additional behavioral predictions that we confirmed in human behavioral data. Finally, we explore the consequences of changing the relative allocation of cognitive resources to the attended versus the unattended item, and show that decision performance is benefited by a more balanced spread of cognitive resources.

Visual attention modulates the integration of goal-relevant evidence and not value

“…Further work in which the size of the set is increased would be required to test this idea more stringently. Notably, two recent pre-prints have also introduced models to explain how the attentional patterns in choice are generated assuming optimal information sampling ( Jang et al, 2020 ; Callaway et al, 2020 ). Both models are based on Bayesian updates of value beliefs, with visual attention playing a role in selecting the information to sample.…”

Trading Mental Effort for Confidence in the Metacognitive Control of Value-Based Decision-Making

“…For example, the efficacy of allocating attention may depend upon which option is considered. In turn, the brain may dynamically refocus its attention on maximally-uncertain options when prospective information gains exceed switch costs (Callaway et al, 2019; Jang et al, 2020). Such optimal adjustment of divided attention might eventually explain systematic decision biases and shortened response times for “default” choices (Lopez-Persem et al, 2016).…”

Fixation patterns in simple choice reflect optimal information sampling

“…First, our results highlight an important role of prior beliefs in sequential sampling models of simple choice (c.f. [ 68 ]). All previous models have assumed a prior mean of zero, either explicitly [ 21 , 68 ] or implicitly [ 9 , 10 , 20 ].…”

“…[ 68 ]). All previous models have assumed a prior mean of zero, either explicitly [ 21 , 68 ] or implicitly [ 9 , 10 , 20 ]. Such a prior is negatively biased when all or most items have positive value, as is often the case in experimental settings.…”

“…In a second line of work, concurrent to our own, Jang, Sharma, and Drugowitsch [ 68 ] develop a model of optimal information sampling for binary choice with the same Bayesian structure as our model and compare their predictions to human behavior in the same binary choice dataset that we use [ 9 ]. There are three important differences between the studies.…”